OpenAI adds Trusted Contact to ChatGPT after 33 AI-linked deaths

OpenAI rolled out a Trusted Contact feature for ChatGPT on Wednesday that lets users 18 and older name an adult to be notified if the service flags a conversation for serious self-harm. The company said the feature expands parental controls it introduced for teen accounts in September 2025.

To set up Trusted Contact, a user adds one adult in ChatGPT settings. The nominated person receives an invitation explaining the role and has one week to accept. If the invite is declined or ignored, the user must choose another contact. When automated systems identify a conversation that may involve severe self-harm, ChatGPT informs the user that it may notify the trusted contact and suggests ways the user can reach out personally.

A team of human reviewers then examines the conversation. If reviewers confirm the exchange meets OpenAI’s threshold for serious risk, the company sends a short alert to the trusted contact by email, text or an in-app notification. The alert does not include the user’s messages; it states a general reason for concern and provides a link with guidance on how to talk with someone in distress. OpenAI said human review is completed within an hour. Users can change or remove their trusted contact at any time, and contacts may decline participation.

OpenAI said it developed the feature with input from its Global Physicians Network, which the company describes as more than 260 licensed doctors across 60 countries, and its Expert Council on Well-Being and AI. The American Psychological Association reviewed the rollout. The APA’s CEO, Dr. Arthur Evans, noted that psychological research shows social connection can reduce risk during periods of emotional distress. Georgia Tech professor and council member Dr. Munmun De Choudhury described the feature as a way to help people access real-world support during vulnerable moments.

OpenAI said Trusted Contact is an additional pathway to real-world help and not a replacement for crisis services. ChatGPT will continue to display local crisis hotline numbers and escalate conversations to suggest emergency services when exchanges reach an acute level of danger.

The feature arrives as OpenAI faces lawsuits from families who allege chatbots contributed to relatives’ deaths by suicide. Public trackers have recorded 33 cases linked to AI chatbots between March 2023 and May 2026, with ChatGPT implicated in the majority of those reports. OpenAI noted Trusted Contact is opt-in and that users can operate multiple ChatGPT accounts, which could limit the feature’s reach.

OpenAI said it will continue working with clinicians, researchers and policymakers to refine how its systems respond when users may be in crisis and to improve connections between AI platforms and human support.